List your company
ResearchFormerly WizWand

Fine-grained Image Classification on Oxford-IIIT Pets

97.3AccuracyCAP
6.50830.07953.6577.221Jan 17, 2021Dec 13, 2021Nov 8, 2022Oct 5, 2023Aug 30, 2024Jul 26, 2025Jun 22, 2026
Updated 4mo ago

Evaluation results

MethodLinks
CAPTraining data setup=primary2021.0197.3——
GPipeTraining data setup=primary2021.0195.9——
C-TPTPre-trained Backbone=CLIP-ViT-L/14, Perturbation budget (epsilon)=4.02025.0494.3—0
OPAMTraining data setup=primary + secondary2021.0193.8——
TPTPre-trained Backbone=CLIP-ViT-L/14, Perturbation budget (epsilon)=4.02025.0493.8—0
MTAPre-trained Backbone=CLIP-ViT-L/14, Perturbation budget (epsilon)=4.02025.0493.7—64.9
R-TPTPre-trained Backbone=CLIP-ViT-L/14, Perturbation budget (epsilon)=4.02025.0493.7—72.9
InterActTraining data setup=primary2021.0193.5——
EnsemblePre-trained Backbone=CLIP-ViT-L/14, Perturbation budget (epsilon)=4.02025.0493.4—63.5
CLIPPre-trained Backbone=CLIP-ViT-L/14, Perturbation budget (epsilon)=4.02025.0493.1—0
TL-AttnTraining data setup=primary + secondary2021.0192.5——
NACTraining data setup=primary2021.0191.6——
Clean (Reference)Backbone=CLIP-ViT-B/16, Labeling Strategy=none (clean accuracy)2026.0690.3——
CLIPBackbone=CLIP-ViT-B/16, Epsilon (ε)=4.02025.0488.30—
CLIPBackbone=ViT-B/16, Epsilon (ε)=4/2552026.0588.3—0
TTPBackbone=ViT-B/16, Epsilon (ε)=4/2552026.0588.3—64.7
C-TPTBackbone=CLIP-ViT-B/16, Epsilon (ε)=4.02025.0488.20—
MTABackbone=CLIP-ViT-B/16, Epsilon (ε)=4.02025.048851.8—
MTABackbone=ViT-B/16, Epsilon (ε)=4/2552026.0588—51.8
TPTBackbone=CLIP-ViT-B/16, Epsilon (ε)=4.02025.0487.40—
R-TPTBackbone=CLIP-ViT-B/16, Epsilon (ε)=4.02025.0487.260.2—
R-TPTBackbone=ViT-B/16, Epsilon (ε)=4/2552026.0587.2—60.2
AGCBackbone=ViT-B/16, Epsilon (ε)=4/2552026.0587—96.1
MTABackbone=ViT-B/32, Epsilon (ε)=4/2552026.0586.3—53.6
MTABackbone=CLIP-ViT-B/32, Epsilon (ϵ)=4/255, Defense Category=Test-time Defense2026.0686.3—53.6
EnsembleBackbone=CLIP-ViT-B/16, Epsilon (ε)=4.02025.0486.251.2—
T-VSSBackbone=CLIP-ViT-B/32, Epsilon (ϵ)=4/255, Defense Category=Test-time Defense2026.0685.3—63.2
CLIPPre-trained Backbone=ViT-B/32, Perturbation budget (epsilon)=4.02025.0485.1—0
CLIPBackbone=ViT-B/32, Epsilon (ε)=4/2552026.0585.1—0
CLIPBackbone=CLIP-ViT-B/32, Epsilon (ϵ)=4/255, Defense Category=Baseline2026.0685.1—0
EnsembleBackbone=ViT-B/16, Epsilon (ε)=4/2552026.0585—65.9
TTPBackbone=ViT-B/32, Epsilon (ε)=4/2552026.0584.7—61
TTPBackbone=CLIP-ViT-B/32, Epsilon (ϵ)=4/255, Defense Category=Test-time Defense2026.0684.7—61
R-TPTPre-trained Backbone=ResNet50, Perturbation budget (epsilon)=1.02025.0484.6—74.2
R-TPTPre-trained Backbone=ViT-B/32, Perturbation budget (epsilon)=4.02025.0484.5—55.8
R-TPTBackbone=ViT-B/32, Epsilon (ε)=4/2552026.0584.5—55.8
R-TPTBackbone=CLIP-ViT-B/32, Epsilon (ϵ)=4/255, Defense Category=Test-time Defense2026.0684.5—55.8
AGCBackbone=ViT-B/32, Epsilon (ε)=4/2552026.0584.1—94
CLIPPre-trained Backbone=ResNet50, Perturbation budget (epsilon)=1.02025.0483.6—0
TTCBackbone=ViT-B/32, Epsilon (ε)=4/2552026.0583.5—11.8
TTCBackbone=CLIP-ViT-B/32, Epsilon (ϵ)=4/255, Defense Category=Test-time Defense2026.0683.5—11.8
EnsembleBackbone=ViT-B/32, Epsilon (ε)=4/2552026.0582.3—66.2
TTCBackbone=ViT-B/16, Epsilon (ε)=4/2552026.0582.3—10.4
APT+TeCoAPre-trained Backbone=ResNet50, Perturbation budget (epsilon)=1.02025.0479.3—79
FAREPre-trained Backbone=ViT-B/32, Perturbation budget (epsilon)=4.02025.0476.7—73.8
FAREBackbone=CLIP-ViT-B/32, Epsilon (ϵ)=4/255, Defense Category=Training-time Defense2026.0676.7—73.8
TeCoAPre-trained Backbone=ResNet50, Perturbation budget (epsilon)=1.02025.0476—75.8
EnsembleBackbone=CLIP-ViT-B/32, Epsilon (ϵ)=4/255, Defense Category=Test-time Defense2026.0675—52.5
TeCoAPre-trained Backbone=ViT-B/32, Perturbation budget (epsilon)=4.02025.0466.9—63.7
TeCoABackbone=CLIP-ViT-B/32, Epsilon (ϵ)=4/255, Defense Category=Training-time Defense2026.0666.9—63.7
APT+TeCoAPre-trained Backbone=ViT-B/32, Perturbation budget (epsilon)=4.02025.0466.7—63.9
APT+TeCoABackbone=CLIP-ViT-B/32, Epsilon (ϵ)=4/255, Defense Category=Training-time Defense2026.0666.7—63.9
APTPre-trained Backbone=ResNet50, Perturbation budget (epsilon)=1.02025.0431.9—3.8
APTPre-trained Backbone=ViT-B/32, Perturbation budget (epsilon)=4.02025.0410—0.2
APTBackbone=CLIP-ViT-B/32, Epsilon (ϵ)=4/255, Defense Category=Training-time Defense2026.0610—0.2
CLIPBackbone=CLIP-ViT-B/16, Attack Type=PGD-100, Perturbation Budget (epsilon)=4/255, Labeling Strategy=pseudo-labels2026.06——3.1
DBDBackbone=CLIP-ViT-B/16, Attack Type=PGD-100, Perturbation Budget (epsilon)=4/255, Labeling Strategy=pseudo-labels2026.06——88.6
R-TPTBackbone=CLIP-ViT-B/16, Attack Type=PGD-100, Perturbation Budget (epsilon)=4/255, Labeling Strategy=pseudo-labels2026.06——66.1